Machine learning-based identification of a metabolic cell death gene signature for assessing disease activity and immunological landscape in inflammatory bowel disease.

Inflammatory bowel disease (IBD) is characterized by a refractory, relapsing inflammatory state driven by a multifaceted and poorly understood interplay between host mucosal metabolic disturbances and immune microenvironment dysregulation. To uncover robust non-invasive diagnostic indicators, this study implemented an integrated analytical framework combining advanced machine learning feature selection, penalized logistic regression modeling, and leave-one-dataset-out cross-validation to construct a novel diagnostic signature derived from metabolic cell death-related genes (MCDRGs). Systematic multiple-testing correction and rigorous data harmonization were applied across all independent discovery and validation cohorts to control for potential batch effects. Through this approach, we successfully identified a core three-gene candidate panel comprising indoleamine 2,3-dioxygenase 1 (IDO1), lipocalin 2 (LCN2), and solute carrier family 6 member 14 (SLC6A14). Methodologically, we demonstrated that the initial near-perfect apparent discrimination within the discovery cohort was mathematically attributable to quasi-complete data separation rather than systemic model overfitting. This identified signature exhibited consistent cross-cohort validation and correlated tightly with the coordinated infiltration and functional states of multiple mucosal immune cell subsets. Furthermore, independent clinical assays substantiated the synchronized elevation of these markers during active clinical phases, while orthogonal single-cell RNA-sequencing confirmed their prominent, cell-type-specific enrichment within the myeloid, epithelial, and stromal compartments of inflamed intestinal mucosa. Collectively, the identified MCDRGs, IDO1, LCN2, and SLC6A14, link metabolic dysregulation, immune infiltration, and regulated cell death, offering insights into IBD pathophysiology and providing a transcriptomic candidate signature with exploratory diagnostic potential for IBD stratification.